An Optimal BP Neural Network Track Prediction Method Based on a GA–ACO Hybrid Algorithm

Abstract

Ship position prediction is the key to inland river and sea navigation warning. Maritime traffic control centers, according to ship position monitoring, ship position prediction and early warning, can effectively avoid collisions. However, the prediction accuracy and computational efficiency of the ship’s future position are the key problems to be solved. In this paper, a path prediction model (GA–ACO–BP) combining a genetic algorithm, an ant colony algorithm and a BP neural network is proposed. The model is first used to perform deep pretreatment of raw AIS data, with the main body of the BP neural network as a prediction model, focused on the complementarity between genetic and ant colony algorithms, to determine the ant colony initialization pheromone concentration by the genetic algorithm, design the hybrid genetic–ant colony algorithm, and optimize this to the optimal weight and threshold of the BP neural network, in order to improve the convergence speed and effect of the traditional BP neural network. The test results show that the model greatly improves the fitness of track prediction, with higher accuracy and within a shorter time, and has a certain real-time and extensibility for track prediction of different river segments.

Improved BP neural network based on genetic algorithm

“…Improper settings can lead to slow convergence and susceptibility to local optima during the training process. Consequently, optimizing the selection of the initial thresholds remains a key concern in BP neural network research [10] .…”

Resource allocation in 5G cloud‐RAN using deep reinforcement learning algorithms: A review

“…This helps the agent to focus on short‐term goals while still considering long‐term goals 40 . The selection of the neural network structure relies on the characteristics of the environment and the specific task being performed 41–43 . The network typically consists of multiple layers, with each layer performing a different type of computation 44,45 .…”

“…40 The selection of the neural network structure relies on the characteristics of the environment and the specific task being performed. [41][42][43] The network typically consists of multiple layers, with each layer performing a different type of computation. 44,45 The hidden layers are designed to extract features from the observations that are relevant to the task.…”

Predictive analysis of COVID 19 disease based on mathematical modelling and machine learning techniques

“…Therefore, it is anticipated that the RL [22][23][24][25][26][27][28][29] will offer a fresh approach to solving this issue. Deep learning models [30][31][32][33] are often able to provide better performance, but it is not always necessary or appropriate to use them. First, the state space and action space defined https://doi.org/10.1016/j.comnet.2023.110105 Received 17 February 2023; Received in revised form 23 July 2023; Accepted 12 November 2023 in this paper are small and simple, so basic reinforcement learning methods may be sufficient to achieve good performance.…”